Banks’ risk models treat the past as the limit of what can happen

Evidence

What VaR does is generate a figure (a VaR number) for how much a firm can win or lose on an individual trade. … What was the worst that could have happened to the US housing market in 2008? As in the height example, the answer depends on a data sample that calibrates the model. Prior to 2007, the worst downturn firms had data on was the result of the mortgage defaults in Texas in the 1980s, when houses lost 40 percent of their value. Take this data as the parameter limit, or how far out the tail goes before the sigma becomes too large to be imaginable, and you will conclude, given the assumption of a normal distribution of events, that the probability that all the mortgages in your portfolio would lose more than 40 percent of their value at once is ridiculously small. — Mark Blyth, Austerity, 109

Categorization

• DEFINE: Explains VaR and the parameter-limit concept — tail calibration determines what models deem possible.
• XREF: Connects to Taleb’s Black Swan critique and the broader financial-crisis literature on model risk the reader is stockpiling. — connects to: Banks blamed the 2008 crash on impossible statistical outlier events — Mark Blyth, Austerity: The History of a Dangerous…, 103; The bell-curve shape of Gaussian distributions is what banks’ risk models assume